[x264-devel] Re: encoding with lots of small slices

[Michael Roitzsch]

Hi x264 team,

> In the meantime, I will try digging a little deeper into the  
> problem and I will get back here if I make any progress.

I think I have found and fixed the first bug concerning slices not  
starting at column one. It's a problem of wrongly selected spatial  
prediction modes with "wrongly" meaning encoder and decoder do not  
agree.

You find two patches attached: The first lists my relaxation of the  
slice border placement so it can place them at arbitrary macroblocks.  
The second patch is the actual fix. If you compare the encoder  
behavior with what the decoder does (x264/decoder/macroblock.c, lines  
731 and 771) the difference should be obvious.

This change removes some of the errors I see in the encoded video.  
The video appears to be error-free when encoding without B-slices. My  
encoder command line is:
x264 --qp 20 --threads 44 --ref 15 --mixed-refs --8x8dct --analyse  
all --direct none -o test.h264 test.yuv 1280x720

With B-slices, there is still something wrong. I will keep trying.

It would be nice if someone could comment on this patch (the second  
one that is) and, if you think it's OK, commit it to SVN if you want.

Michael

-------------- next part --------------
A non-text attachment was scrubbed...
Name: relax-slice-tests.patch
Type: application/octet-stream
Size: 1228 bytes
Desc: not available
Url : http://mailman.videolan.org/pipermail/x264-devel/attachments/20060725/04c0c117/attachment.obj 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: spatial-prediction-fix.patch
Type: application/octet-stream
Size: 673 bytes
Desc: not available
Url : http://mailman.videolan.org/pipermail/x264-devel/attachments/20060725/04c0c117/attachment-0001.obj 
-------------- next part --------------